sides a total budget of 150 ms can be consumed
without hampering the quality. However, from
Table 3 it can be seen that if the performance of
the echo control reduces to “nearly perfect”, but
still is standard-compliant, the bound on the
mean one-way mouth-to-ear delay can be below
150 ms in some cases. The packetization delay
is typically chosen between 10 and 80 ms. From
eq. (1) it follows that since the overhead SOHis
320 bits (consisting of 20 IP, 8 UDP and 12 RTP
bytes) for Voice over IP (VoIP), an overhead bit
rate between 32 kb/s and 4 kb/s, respectively, is
introduced.
The flexibility in the choice in dejittering loss
Ploss,jit(or equivalently the dejittering delay Tjit)
is governed by the number of quantiles that are
specified in the SLS, i.e. how many points of the
function F(.) are given. If only the maximum
total queuing delay (i.e. the (1-P)-quantile with
P= 0) is given, only this total queuing delay can
be used as dejittering delay. The more quantiles
are specified, the more flexible the choice can be.
To conclude this section we give an example.
Consider a phone call from Europe to the US.
We assume Tnet,min= 50 ms, Ploss,net= 0 and
that the SLS in both directions is as described
in Table 5(a). Furthermore, we assume a DSP
delay TDSP= 15 ms, an echo loss EL 1 =EL 2
= 50 dB, and that the G.729 codec (at 8 kb/s) is
used. Table 5(b) gives the effective bit rate Reff
calculated with eq. (1) and Table 5(c) (using the
color code of Table 1) gives the rating Rcalcu-
lated with eq. (10) for various values of the
packetization delay and dejittering loss. From
these tables it can be concluded that a packetiza-
tion delay of 30 ms and a dejittering loss of 10-3
lead to a good compromise between effective bit
rate (Reff= 18.7 kb/s) and quality (R= 79).
The question how to provision the SLSs, i.e.
how to configure the routers in the network
such that the quantiles specified in the SLS are
attained is beyond the scope of this paper. We
refer the interested reader to [11] and [17].
5 Conclusions
In this paper the quality issues associated with
the packetized transport of phone calls were con-
sidered. Since for packetized phone calls more
delay and distortion is introduced than for tradi-
tional PSTN calls, the impact of delay and dis-
tortion on the quality of the phone call was stud-
ied quantitatively with the E-model. The trade-
offs involved in the choice of the packetization
delay and dejittering loss were discussed. From
this quality study the following conclusions were
drawn.
For packetized phone calls echo control is highly
recommended, if not required, since otherwise
the tolerable mouth-to-ear delay budget risks
being too small. If the echo is perfectly con-
trolled, the quality remains equal to the intrinsic
quality up to a mouth-to-ear delay of about
150 ms. The intrinsic quality depends on the
amount of distortion that is introduced. If the
echo control is slightly less than perfect, but still
standard-compliant, the quality decreases even
for delays smaller than 150 ms.The intrinsic quality associated with predictive
codecs at low bit rates is lower than the tradi-
tional PSTN quality. Therefore, these codecs
should not be used at a bit rate below 32 kb/s.
For the same reason, transcoding should be
avoided.Under perfect echo control the margin between
the intrinsic quality of a codec and the bound for
traditional quality can either be consumed by
allowing a mouth-to-ear delay above 150 ms or
by allowing some packet loss. The maximum
tolerable bounds on the mean one-way mouth-
to-ear delay and packet loss are reported in this
paper for the most common codecs and even the
recently developed Adaptive MultiRate (AMR)
codec. It is also shown how these bounds de-
crease if the echo control is slightly less than
perfect, but still standard-compliant.These tolerable bounds should be respected by
any packetized phone call (gateway-to-gateway,
IP-phone-to-IP-phone, mobile-phone-to-mobile-
phone, gateway-to-IP-phone, etc.) if traditional
quality is to be maintained.Finally, to illustrate how these bounds can be
used this paper considered a gateway-to-gateway
scenario where the transport of the voice packets
is governed by a Service Level Specification
(SLS). The trade-offs involved were shown by
means of a numerical example.Acknowledgement
This work was carried out within the framework
of the project LIMSON sponsored by the Flem-
ish Institute for the Promotion of Scientific and
Technological Research in the Industry (IWT).